采用一種新的基于獨立分量分析（independent component analysis, ICA）盲源分離（blind source separation, BSS）和形態學開重構（Open Reconstruction）的方法實現多極化合成孔徑雷達（synthetic aperture radar, SAR）影像相干斑噪聲抑制和水體目標快速提取。SAR影像具有強烈乘性相干斑噪聲，影像數據為非高斯分布，具體分布形式未知。利用獨立分量分析方法，不需要知道SAR影像的具體分布，通過對數量化將相干斑噪聲轉化為與圖像數據相互獨立的加性噪聲，從多極化SAR影像中自動分離出圖像數據與相干斑噪聲，并自動選擇相干斑指數最小的分量為圖像分量。針對SAR影像水體目標的亮度及形狀分布特征，進一步采用形態學開重構運算，從分離出的圖像分量中提取出水體目標。利用ENVISAT ASAR多極化影像進行了實驗，結果表明該方法可以快速準確地提取多極化SAR影像中的水體目標。

關鍵詞：合成孔徑雷達; 水體目標; 自動識別; 盲源分離; 獨立分量分析; 數學形態學; 開重構

Method for water object extraction in SAR imagery based on morphology and blind source separation
WANG Dong1, CHEN Ying-ying1,2, QIN Ping3
1. Dept. of Surveying and Geo-informatics, Tongji University, Shanghai 200092, China
2. Research Center for Remote Sensing and Spatial Information, Tongji University, Shanghai 200092, China
3. East China Sea Marine Prediction Center, State Bureau of Oceanic Administration, Shanghai 200082, China

Abstract:?A new method is proposed for speckle noise suppression and water objects extracting from synthetic aperture radar(SAR) imagery based on morphology and independent component analysis(ICA) blind source separation(BSS). The distribution of SAR image data with multiplicative speckle noise is non-Gausssian and its parameters are unknown. Logarithmic Quantification is utilized to transform multiplicative speckle noise to independent additive noise. Speckle noise and image data are separated from multi-polarimetric imagery, and the components with the least speckle index are chosen as the object component automatically by means of ICA while the specific distribution of SAR imagery is unnecessary. Water objects are extracted from the separated component imagery based on Morphology Open Reconstruction according to their lightness and region shape features. Experimental results for ENVISAT ASAR show that the method can extract water objects from the multi-polarimetric imagery with high accuracy and fast speed.
Key words:?synthetic aperture radar; water target; automatic recognition; blind source separation; independent component analysis; mathematical morphology; open reconstruction